Who has a Barton?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
Who has a Barton?
Edit: I have what I need. Thank You to all who participated in my poll.
If you have an AMD Athlon "Barton" cpu, please contact me, either by replying to this thread or PM'ing me. Email is fine too, but I don't have access to that at work, so a PM is probably better.
David
If you have an AMD Athlon "Barton" cpu, please contact me, either by replying to this thread or PM'ing me. Email is fine too, but I don't have access to that at work, so a PM is probably better.
David
Last edited by haysdb on Mon Apr 26, 2004 10:03 pm, edited 1 time in total.
-
- Posts: 5
- Joined: Sun Apr 18, 2004 12:35 pm
- Location: Pennsylvania (USA)
- Contact:
-
- Posts: 3
- Joined: Tue Mar 30, 2004 9:04 pm
- Location: San Luis Obispo Ca. usa
Oh, I might as well just fess up. This cloak and dagger stuff just isn't my style.
What I have observed is that my Bartons seem to be doing exceptionally well on DGromacs. Not as well as my P4's (which have SSE2), but better than they do on Tinkers, which are sometimes pretty good, sometimes not so good. There are still some discrepancies in the points values of various Tinker units IMO.
The "secret" which I was sharing with fellow Barton users was to continue using -advmethods, since this seems to be the only way to get DGromacs WU's on non SSE2 processors.
The reason I specifically say "Barton" rather than "Athlon" is that it may be that the Bartons do better on the DGromacs than the T'breds and Palomino's, but this may not be true at all, I just don't have enough experience with DGromacs on T'breds to say. I have had a few units on Palomino's and they don't seem to like them as well as the Bartons.
Since Bartons and T'breds are similar cores, differing only in the amount of cache, the only reason the Barton's might do better is that the DGromacs are able to take advantage of the extra cache, whereas Tinkers and regular Gromacs do not seem to utilize the extra cache of the Bartons. Pure conjecture here folks, and possibly pure BS.
Again, this is preliminary and rather "anecdotal", so don't take any of this as gospel, and I would appreciate feedback on everyone's own experiences with DGromacs on Athlon processors.
In summary:
-advmethods is still recommended on P4 systems, with no caveats.
The efficacy of -advmethods on Athon's is still open to debate. With it, you may get more DGromacs, which may be worth more PPD than Tinkers. Without it you may get more Tinkers, which may be worth more points, and may be worth less points, than regular Gromacs. It's a crap shoot.
David
What I have observed is that my Bartons seem to be doing exceptionally well on DGromacs. Not as well as my P4's (which have SSE2), but better than they do on Tinkers, which are sometimes pretty good, sometimes not so good. There are still some discrepancies in the points values of various Tinker units IMO.
The "secret" which I was sharing with fellow Barton users was to continue using -advmethods, since this seems to be the only way to get DGromacs WU's on non SSE2 processors.
The reason I specifically say "Barton" rather than "Athlon" is that it may be that the Bartons do better on the DGromacs than the T'breds and Palomino's, but this may not be true at all, I just don't have enough experience with DGromacs on T'breds to say. I have had a few units on Palomino's and they don't seem to like them as well as the Bartons.
Since Bartons and T'breds are similar cores, differing only in the amount of cache, the only reason the Barton's might do better is that the DGromacs are able to take advantage of the extra cache, whereas Tinkers and regular Gromacs do not seem to utilize the extra cache of the Bartons. Pure conjecture here folks, and possibly pure BS.
Again, this is preliminary and rather "anecdotal", so don't take any of this as gospel, and I would appreciate feedback on everyone's own experiences with DGromacs on Athlon processors.
In summary:
-advmethods is still recommended on P4 systems, with no caveats.
The efficacy of -advmethods on Athon's is still open to debate. With it, you may get more DGromacs, which may be worth more PPD than Tinkers. Without it you may get more Tinkers, which may be worth more points, and may be worth less points, than regular Gromacs. It's a crap shoot.
David
i tried that on my server which is a tbred 2000+ and the barton runs circles around it. i got a wu on each machine that was 10000 points. the barton at 1.86 compleated the wu almost 40% faster then the tbred at about 1.75 ghz. worth further study? i removed the swich from my server to get maximum pointage!!
I must be missing something here, but I have yet to see the great speed increase with Tinkers on my two Bartons (each has gotten 3-4 Tinkers since I removed the flag) -- they have been doing the exact same PPW as Gros on my boxes.
My experiment is over, I'm going back to -advmethods with the Bartons, they don't take as long and I may get a Dgro once in awhile...
Dave
My experiment is over, I'm going back to -advmethods with the Bartons, they don't take as long and I may get a Dgro once in awhile...
Dave
Some of the Tinkers do quite well, points wise, but certainly not all of them. I have seen over 1K PPW, and I have seen 500, on the same systems. I'd say ON AVERAGE, I'm not scoring any better on Tinkers than I am on Gromacs, but not really any worse either. So, if it weren't for the DGromacs, I'd say it doesn't really matter, purely in terms of points, whether -advmethods is used or not used. The DGromacs tip the scales in favor of continuing to use -advmethods.
David
David
-
- Posts: 237
- Joined: Tue Nov 11, 2003 6:48 am
- Location: Gothenburg, Sweden
- Contact: